Query-Focused Association Rule Mining for Information Retrieval

We present a method that applies association rule mining for information retrieval. Our approach is different from traditional information retrieval since retrieval is done based on association rather than similarity, which might be useful for knowledge discovery purposes such as finding an explanation or elaboration for an event in a collection of domain-specific documents. The method proposed in this paper is based on the SmoothApriori algorithm which accommodates similarity in the association rule mining process to mine association rules between sentences or larger text units. We introduce query-focused association rule mining that allows association-based retrieval from larger amount of data than with a traditional association-rule mining approach. Combined with SmoothApriori, query-focused association rule mining provides association-based retrieval for textual data. This new method was evaluated on the task of automatically restoring sentences that were artificially removed from aviation investigation reports and showed significantly better results than any of our similarity-based retrieval baselines.